Adaptive Image Compression Using Sparse Dictionaries

Size: px
Start display at page:

Download "Adaptive Image Compression Using Sparse Dictionaries"

Transcription

1 Adaptive Image Compression Using Sparse Dictionaries Inbal Horev, Ori Bryt and Ron Rubinstein Abstract Transform-based coding is a widely used image compression technique, where entropy reduction is achieved by decomposing the image over a dictionary known to provide compaction. Existing algorithms, such as JPEG and JPEG2000, utilize fixed dictionaries which are shared by the encoder and decoder. Recently, algorithms using content-adaptive dictionaries have been proposed. These algorithms employ more specialized dictionaries which are optimized for specific classes of images, and in this way are able to improve compression rates for these classes. Such approaches, however, lose generality as they require sharing a specialized dictionary in advance between the encoder and decoder, for every type of input image. Utilizing image-adaptive dictionaries has the potential of restoring generality and improving compression rates by encoding any given input image over a dictionary specifically adapted to it. However, this has so far been avoided as it requires transmitting the dictionary along with the compressed data. In this work we use the sparse dictionary structure to implement generic image compression. This dictionary structure has a compact representation, and thus can be transmitted with relatively low overhead. We employ this structure in a compression scheme which adaptively trains the dictionary for the input image. Our results show that although this method involves transmitting the dictionary, it remains competitive with fixed-dictionary schemes such as JPEG and JPEG2000. keywords: Image compression, image coding, sparse representation, dictionary learning, sparse K-SVD. The Signal and Image Processing Lab, Electrical Engineering Department, The Technion - Israel Institute of Technology, Technion City, Haifa 32000, Israel. The Geometric Image Processing Lab, Computer Science Department, The Technion - Israel Institute of Technology, Technion City, Haifa 32000, Israel, ronrubin@cs.technion.ac.il. 1

2 1 Introduction Compression of natural images relies on the ability to capture and exploit redundancies found in these images. The most common compression approach, known as transform coding, utilizes a dictionary of atomic signals, such as the DCT or wavelet dictionaries, over which the image is known to be compressible. The dictionary is typically arranged as a matrix D = [d 1 d 2... d L ] R N L, with the columns d i constituting the atoms, and L N. Given a signal x R N, compression is achieved by approximating it as a linear combination of the atoms, x Dγ, (1) where the representation vector γ is expected to have lower entropy than the entries of x. When D is invertible, the representation γ can be computed by inverting D and quantizing the coefficients: γ = Q(D 1 x). This is the case in the JPEG [1] and JPEG2000 [2] compression standards, where D is the DCT or wavelet dictionary, respectively. When D is overcomplete (L N), the null space of D introduces additional degrees of freedom in the choice of γ, which can be exploited to improve its compressibility. The representation is typically selected by minimizing some penalty function C(γ) which estimates its compressibility, such as the l 0 penalty C(γ) = γ 0 which measures the number of non-zeros in the representation: ˆγ = Argmin γ γ 0 Subject To x Dγ 2 2 ɛ 2. (2) Here, ɛ is the approximation error target, controlling the distortion of the compressed signal. This problem is known as the sparse approximation problem [3], and though NP-hard in general, it can be approximated by a wide range of techniques [4, 5, 6]. Other choices for C(γ) include the variety of robust penalty functions such as the l p cost functions with 0 p 1. All these functions promote the sparsity of the representation γ (i.e., the fast decay of its coefficients) by strongly penalizing small non-zero values. We should remark that for a practical compression scheme, compressibility of a representation is affected by additional factors other than sparsity (e.g. quantization, entropy coding etc.). Nonetheless, sparsity provides a simple and relatively reliable approximation of compressibility. Transform-based coding schemes generally assume the dictionary D to be fixed, and built into both the encoder and decoder. This is the case for the JPEG family of algorithms, which are based on predetermined fixed dictionaries and are targeted at general-purpose image compression. Recently, compression schemes aimed at more specific classes of images have been developed, and show substantial gains by employing a content adapted dictionary which is optimized for a specific class of images [7, 8, 9]. In this work, we take a different approach and target general image compression using adaptive dictionaries. Our goal is to increase sparsity by encoding the image over 2

3 a specifically-trained dictionary adapted for the input image. This goal is ambitious, as it requires transmitting the dictionary along with the compressed data, which introduces significant overhead. To address this, we propose using a parametric dictionary, which can be represented by a relatively small number of values. Several such dictionaries have been recently proposed (see [10]). Our work focuses on the sparse dictionary structure [11], which was chosen due to its simplicity and ability to represent relatively rich dictionaries. Our compression scheme trains the dictionary specifically for the input image, and encodes it as part of the compressed stream. In this way, the compression method can accommodate a wide range of images, since it imposes fewer assumptions on their behavior. Our simulations show that even though our method must encode the dictionary as part of the stream, it consistently outperforms JPEG compression, and comes close to JPEG2000 in several cases. We view these as significant and encouraging results, which demonstrate the feasibility of the image-adaptive approach, and open the door to further research. 1.1 Related Work Several works on compression using content-adaptive dictionaries have been recently published. In all these works, the trained dictionary is assumed to be known to both the encoder and decoder in advance. One of the first works to successfully employ this approach is [7], where the authors propose an algorithm for facial image compression. The algorithm employs a pre-processing geometric alignment step, followed by a sparse approximation of the image patches over a set of pre-trained dictionaries. The method is shown to achieve dramatically higher rates than JPEG and JPEG2000 for facial imagery due to the optimized dictionaries, and clearly demonstrates the potential of content-aware compression. The approach, however, is not immediately extendible to more complex classes of images. A different method, applicable to a wider range of image families, is proposed in [8]. In this work, a set of orthogonal dictionaries is pre-trained for a given class of images, and the compression is implemented by allowing each patch in the input image to select its optimal dictionary from the set. The authors show that for facial images, this method typically outperforms JPEG and comes close to JPEG2000. Experiments with natural images show varying performance, matching or surpassing JPEG. Finally, a method based on iteration-tuned dictionaries (ITDs) has been recently proposed [9]. In this work, a single hierarchical ITD is pre-trained for the family of images, and used to encode the input image patches. The authors test their method with facial images, and show that it can convincingly outperform JPEG and JPEG2000 for this class of images. Other classes of images remain to be investigated. 3

4 Figure 1: Left: overcomplete dictionary for 8 8 image patches, trained using the K-SVD algorithm. Right: images used for the training. 1.2 Report Organization This report is organized as follows: In section 2 we review the sparse dictionary structure, which forms the core of our algorithm. The compression scheme is described in section 3, followed by results in section 4. We conclude in section 5, where we also discuss a few future research directions. 2 Sparse Dictionaries The sparse dictionary structure is a parametric dictionary model recently proposed as a means of bridging the gap between analytic and trained dictionaries [11]. It is a simple and effective structure based on sparsity of the atoms over a known base dictionary. The motivation for this structure comes from the observation that dictionaries trained from natural image data are typically highly structured, and show notable regularity. For example, Fig. 1 shows a dictionary trained using the K-SVD algorithm [12] on a set of 8 8 natural image patches. The regularity of the trained atoms suggests that these atoms themselves may have some underlying sparse structure over a more fundamental base dictionary. Thus, according to this view, the dictionaries describing different images may not be completely arbitrary, and instead have a common underlying explanation in the form of the base dictionary. This base dictionary in turn consists of a set of signals which can be thought of as sub-atomic signals from which all observable dictionary atoms are formed. Formally, the sparse dictionary structure represents each atom of the dictionary as a sparse combination of atoms from a prespecified base dictionary Φ. The dictionary is 4

5 therefore expressed as D = ΦA, (3) where A is the atom representation matrix, and is assumed to be sparse. For simplicity, we assume A has a fixed number of non-zeros per column, so a i 0 p for some p. The base dictionary Φ is a fixed non-adaptive dictionary which is part of the model. Benefits of this model include adaptability (via modification of A), efficiency (assuming Φ has an efficient implementation), and compact representation (as only A requires specification). Training the sparse dictionary is done using the Sparse K-SVD algorithm [11], which efficiently adapts the matrix A given a set of examples. The algorithm alternates between sparse coding and dictionary update steps, taking a similar form to the original K-SVD algorithm [12]. We refer the reader to [11] for a complete description of the algorithm. 3 Adaptive Image Compression The adaptive encoding process is summarized in Fig. 2. The process begins by partitioning the image to non-overlapping patches and subtracting the mean (DC) value from each. The DC values are subsequently quantized, and their running differences are entropy coded. The DC-free patches, which contain the bulk of the image information, are used to train a sparse dictionary using Sparse K-SVD. As the base dictionary, we use the overcomplete DCT 1, which is known to be an efficient generic dictionary for representing small image patches. The outcome of this training is a matrix A describing an image-specific dictionary for representing the image patches. This matrix undergoes quantization and is then used to encode the DC-free patches. We perform sparse coding over the quantized dictionary D q = ΦA q to allow inversion of the process at the decoder. For the sparse coding, we use a variant of Orthogonal Matching Pursuit (OMP) [5] which we name Global OMP. The sparse coding step produces a sparse matrix Γ with the sparse representations of the patches as its columns, and Γ is subsequently quantized to form Γ q. Finally, both A q and Γ q are fed to a sparse matrix encoder which generates the compressed representation of the DC-free content. The full compressed stream consists of the encoded DC values and the two compressed sparse matrices. We note that the sparse dictionary is learned from zero-mean patches, however the sparse coding step is performed with patches from which quantized DC values were subtracted, and therefore may have non-zero means. We resolve this discrepancy by adding a fixed constant-valued (DC) atom to the trained dictionary, and implicitly assume its existence as the first atom of the dictionary in both the encoder and decoder. Decoding the stream is straightforward and efficient, and includes reversing the sparse matrix 1 The overcomplete DCT dictionary is an extension of the standard DCT dictionary which allows non-integer wave numbers. 5

6 Image Partition to Blocks DC Removal DC Values DC Encoder DC enc Multiplexer Compressed Stream DC-Free Blocks A enc, Γ enc Sparse K-SVD A Q A q Sparse Coding Γ Q Γ q Sparse Matrix Encoder Figure 2: The proposed encoding scheme. encoding, computing the DC-free patches X = ΦAΓ, and restoring the encoded DC values. In the next sections we describe in more detail the key components of the scheme. 3.1 Global OMP Our implementation of the compression scheme accepts a target PSNR as the control of the output rate. In the sparse coding stage, this target can be enforced individually for each image patch by setting ɛ in (2) to ɛ 2 = Imaxb/10 2 P SNR 10, where b denotes the number of pixels in a patch. Alternatively, we can lift this constraint and allow the error to be distributed arbitrarily among the patches. This results in a more flexible sparse coding scheme which potentially achieves higher sparsity. Thus, we solve a global sparse coding problem for all image patches simultaneously: Min Γ Γ 0 Subject To Y D q Γ 2 F ɛ 2 g. (4) Here, Y is a matrix with the image patches as its columns, and ɛ g is the global error target for the image. We note that a similar process was independently developed and used in [13]. Solving problem (4) is equivalent to sparse coding a column-stack representation y of Y over the dictionary I D q (the block matrix containing instances of D q as its main diagonal): Min γ 0 Subject To y (I D q )γ 2 γ 2 ɛ 2 g. (5) The resulting sparse coding problem is solved using OMP, which can be implemented efficiently through localization of the computations to a single patch each iteration. Specifically, we store for each patch its current residual vector, sparse coefficients, and inner-products with the dictionary atoms. At the selection stage, we choose the largest inner product among all image patches, which determines both the patch to process and the atom to add to its representation. The addition of the atom involves only local updates to the patch information, and thus introduce no overhead compared to encoding the patch independently. We name the resulting process Global OMP, as it globally processes all image patches towards a single collective error target. We use it in the dictionary training as well as the sparse coding steps, in order to better tune the learned result to the global process. 6

7 3.2 Quantization We quantize the non-zero values in A and Γ using a uniform quantizer. While the distribution of these values is highly non-uniform, it is known that using uniform quantization followed by entropy coding generally outperforms non-uniform quantization. A side effect of the quantization of Γ is that the PSNR target achieved by the sparse coding step is lost. To restore the desired PSNR target, we employ a simple iterative refinement process, in which coefficients are added to Γ to compensate for the quality loss. We begin with the original PSNR target and its associated error value ɛ g. We denote this original (user-specified) PSNR target by p 0 and the achieved PSNR after quantization by q 0 (< p 0 ), and let r 0 = p 0 q 0. Assuming r 0 is relatively small, we estimate that the quantization-induced PSNR loss will be approximately the same for any target PSNR close to p 0. This implies that the target PSNR p 1 = p 0 + r 0 should roughly achieve the user-specified PSNR target after quantization. We therefore add coefficients to Γ until reaching the updated target p 1, by continuing the greedy Global OMP from the point it terminated at p 0. After quantization, if the result is still below the desired PSNR target, the Global OMP target can be raised again using a similar process, based on the updated PSNR loss r 1 = p 1 q 1. The process repeats as necessary until reaching the user-specified PSNR target p 0 ± δ for some δ. The overall process is efficient and requires relatively few repetitions (typically 2-5). It should be noted that since the Global OMP is continued rather than restarted, the overhead of these repetitions (compared to having known the correct sparse coding target to begin with) is small. 3.3 Sparse Matrix Encoding Our sparse matrix encoder represents the matrices A q and Γ q in column compressed (CCS) form. It encodes the values of the coefficients via quantization and entropy coding, and the locations of these coefficients via difference coding of the row indices followed by entropy coding. A useful observation is that the order of the columns in A is arbitrary, and is essentially a degree of freedom of the representation. Indeed, we can apply any permutation to the columns of A, along with the same permutation to the rows of Γ, without altering the product AΓ. This freedom can be used to improve the compressibility of the row indices in Γ. In this work we reorder the columns of A such that they become ordered in decreasing order of popularity. In other words, the rows of Γ are sorted in descending order of non-zero count. This sorting results in concentration of the non-zero values in Γ near the top of the matrix, and thus the overall entropy of the index differences is reduced. To facilitate the difference-coding of the row indices in Γ, we transmit for each column the index of the first non-zero value, followed by a sequence of differences for the remaining indices. Owing to the above sorting, the order of the rows in Γ is such that the index of the first non-zero value in each column is typically small, and hence has a 7

8 particulary low entropy. We thus use a dedicated entropy coder for the indices of the first value in each column, and a separate entropy coder for the index differences. We indicate an empty column in Γ by sending the special symbol 0 as the index of the first non-zero value in that column (we index valid matrix rows from 1). For a non-empty column which can have a variable number of non-zero values we indicate the end of its index sequence by sending the special symbol 0 as the index difference. For the coefficient values, we entropy-code their absolute value as a consecutive stream (a different stream for A and Γ). The signs are sent unprocessed. 3.4 Entropy Coding The entropy coding in this work is implemented using an arithmetic coder. We recall that given a set of symbols, the arithmetic coder and decoder require the symbol probabilities {p i } as side information. These probabilities are determined by the encoder, and must be transmitted to the decoder. To avoid sending floating-point numbers, we quantize and transmit the log-probabilities log 2 (1/p i ). These values represent the optimal codeword lengths of the symbols, and thus have a relatively small range which can be uniformly quantized. We have found that using very few bits (5 6) for the quantized values results in practically no increase to the code length, while providing an effective way of transmitting the side information. 3.5 Parameter Tuning One can imagine that compression schemes, such as the one described in [7], rely on many parameters that need to be set before actual coding. Some of these parameters can be predetermined, while others depend on the image content and the requested output quality. The proposed scheme involves several such parameters as well. In this section we discuss the main parameters in the scheme the patch size, the dictionary size, the atom sparsity and the quantization step sizes and their effect on the compression rate. Our experiments have shown that of the above parameters, only a few have a critical effect on compression performance. Based on these experiments, our system implements a semi-automatic parameter tuning process which requires no manual intervention. We should remark that although this heuristic process has been found relatively effective, manual experimentation has verified that it is still sub-optimal, and improved results can be achieved by further refining it. For the dictionary size and atom sparsity, our system employs hard-coded values which depend only on the patch size, as detailed in Table 1. Based on our experiments, we have found that the optimal values for these parameters are quite consistent among images assigned with the same patch sizes, with compression results remaining stable when deviating from these values. Thus, fixed values for these two parameters suffice for our encoding system. In the same way, our base dictionary size is fixed as well, and depends on the patch size. For a patch size of N N, our overcomplete DCT base 8

9 Patch Size Dictionary Size Atom Sparsity Table 1: Dictionary size and atom sparsity for each patch size. dictionary is of size N 2 (N + 2) 2, i.e., the Kronecker product of two 1-D N (N + 2) overcomplete DCT dictionaries. Selecting the quantization steps is a more elaborate process. Our scheme involves three quantization steps which must be chosen: one for the DC values, and two for the non-zero values in A and Γ. Beginning with the DC quantization step, we recall that our scheme adds a fixed DC atom to the trained dictionary to overcome DC quantization effects. Thus, coarser DC quantization results in more non-zero values appearing in Γ, related to this DC atom. Our selection rule for the DC quantization step heuristically chooses this value such that the increase in non-zero count in Γ due to the DC quantization is around 6%. Regarding the quantization steps for the non-zeros in A and Γ, we notice that for both these values there is a direct trade-off between the harshness of the quantization and the number of coefficients that will be required in Γ to achieve a given PSNR target. Specifically, by coarsening the quantization, it remains possible (up to some point) to satisfy the PSNR target, but at the expense of more non-zero values added to Γ. Our system employs a heuristic process which simultaneously selects both step sizes, with the goal of achieving a 0.85dB PSNR loss due to the quantization. This loss is then compensated for by adding coefficients to Γ. To achieve this goal, we extend the iterative process in 3.2 to repeatedly refine both step sizes based on the current PSNR loss, modifying the step sizes as necessary according to a set of empirically designed rules. Finally, we have found that one of the most important parameters controlling compression performance is the patch size. Unfortunately, we have not yet found a sufficiently effective heuristic for selecting this size. Thus, our system determines the optimal size for each image by applying several predetermined patch sizes in a highly reduced compression scheme, and choosing the size that achieves the optimal rate for the target PSNR. 9

10 Patch Size PSNR Target [db] Figure 3: The effect of the patch size choice on the compressed file size, for the image Barbara. For each PSNR target, the corresponding column shows the ratios between the obtained file sizes for different patch size choices, and the optimal file size for that PSNR. We consider patch sizes from 3 3 pixels to 14 14, as listed in Table 1. As expected, we have found that lower PSNR targets generally prefer larger patch sizes, due to the reduced accuracy required by the compression. Fig. 3 illustrates the relation between the patch size and the resulting file size for the image Barbara. For each PSNR target, the figure shows the relative increase in file size incurred by different patch size choices, compared to the optimal size. As can be seen, compression performance is quite stable under minor deviations from the optimal patch size, though selecting an unsuitable size can result in substantially larger files. For additional details on the parameter tuning process, we refer the reader to [14]. 4 Results We have tested the proposed scheme on a variety of images, and the results for seven standard test images are presented below. The Sparse K-SVD results were produced using the parameter selection process described in the previous section. The JPEG and JPEG2000 images were produced using MATLAB R2010a. Representative compression results are listed in Table 2. Figs. 4-6 show the corresponding compressed images, and Fig. 7 presents comparative rate-distortion graphs. The three parts in the table correspond to the three figures 4,5,6. In each part, all test images are compressed to the same target PSNR by the Sparse K-SVD compression scheme, and the JPEG and JPEG2000 algorithms are tuned to match the resulting file size. The three cases represent low, medium and high bit-rates, corresponding to PSNR targets of 25dB, 29dB and 34dB for the Sparse K-SVD. The rate-distortion graphs summarize our complete results for the seven test images. As can be seen, our scheme consistently outperforms JPEG, and comes close to 10

11 JPEG2000 in several cases. Our method typically performs better on images containing more texture, owing to the ability of the dictionary to capture and efficiently represent repetitive behavior. Similar to the JPEG algorithm, our method suffers from blockiness due to partitioning of the image. This artifact can likely be reduced by employing a post-processing deblocking scheme. Finally, Fig. 8 demonstrates a typical decomposition of the compressed stream resulting from our encoder. As can be seen, the indices of the representation coefficients in Γ occupy the majority of the compressed file. This behavior is due to the random structure of the coefficients in Γ, which exhibits little organization or repetitiveness, and is thus difficult to compress efficiently. File Size PSNR (db) Figure Image (KB) JPEG 2 Sparse K-SVD JPEG2000 barbara 5.9 ± lena 3.4 ± peppers 3.9 ± pirate 5.1 ± zentime 5.4 ± table 14.5 ± dollar 23.4 ± barbara 11.1 ± lena 4.77 ± peppers 6.33 ± pirate 13.3 ± zentime 16.3 ± table 30.2 ± dollar 41.9 ± barbara 26.7 ± lena 12.6 ± peppers 14.8 ± pirate 36.6 ± zentime 39.3 ± table 59 ± dollar 72 ± Table 2: Quantitative comparison of JPEG 2, JPEG2000 and Sparse K-SVD compression. Each part of the table corresponds to a single target PSNR for the Sparse K-SVD (25dB, 29dB and 34dB), and the JPEG and JPEG2000 are tuned to match the resulting file size. 2 The empty entries in the table correspond to cases where the Sparse K-SVD file sizes were below the possible minimum of the JPEG algorithm. The smallest file sizes achieved by JPEG for Lena and Peppers were 4.27KB (24.25dB) and 4.42KB (24.3dB), respectively. The corresponding images are shown in Fig

12 Figure 4: Visual comparison of the schemes (low bit-rate). Left to right: Original, JPEG, Sparse K-SVD and JPEG2000. For the images Lena and Peppers, where JPEG could not achieve the target file size, the shown images use the lowest possible quality settings. 12

13 Figure 5: Visual comparison of the schemes (medium bit-rate). Left to right: Original, JPEG, Sparse K-SVD and JPEG

14 Figure 6: Visual comparison of the schemes (high bit-rate). Left to right: Original, JPEG, Sparse K-SVD and JPEG

15 Figure 7: Rate-distortion curves for the seven test images. Comparison of JPEG, Sparse K-SVD, and JPEG2000 compression. 15

16 Figure 8: Decomposition of the compressed stream for the image Dollar. 5 Conclusion and Future Directions This work has presented a new image compression scheme based on image-adaptive dictionaries. The system is unique in that it encodes the image over a dictionary specifically trained for the input. This approach, which requires transmission of the dictionary as part of the compressed stream, is made possible owing to the compact representation of the sparse dictionary structure. We have shown that despite the overhead in sending the dictionary, our system consistently outperforms the JPEG algorithm, which is a similar patch-based scheme, but utilizes a pre-shared fixed dictionary. Indeed, while our current implementation does not reach JPEG2000 performance, our results remain significant in that they demonstrate the feasibility and potential of the adaptive approach. Such an approach, as far as the authors are aware of, has so far been considered impractical. Many enhancements to the scheme could be introduced. Most notably, working with several image scales could more efficiently represent differently-sized features, as well as eliminate the need to select a patch size for each input individually. Alternatively, enabling variable-sized patches based on local image complexity could also accomplish this. Another interesting way to achieve multi-scale behavior is to apply the scheme on the wavelet (or other multi-scale) transform of the image, which could also reduce blockiness effects. In another direction, an important observation is that the encoded indices occupy a significant part of the resulting compressed stream. Thus, discovering hidden patterns in Gamma, or alternatively, modifying the sparse coding process to create more regular patterns, could dramatically improve compression efficiency. Finally, the scheme could be extended to allow fixed pre-shared parts in the dictionary alongside adaptive ones, thus reducing the overall cost of sending the dictionary. 6 Acknowledgements The authors would like to thank Prof. Michael Elad, Prof. David Malah and Nimrod Peleg for their guidance and helpful advice throughout this work. 16

17 References [1] W. B. Pennebaker and J. L. Mitchell, JPEG still image data compression standard. Springer, New York, [2] D. S. Taubman and M. W. Marcellin, JPEG2000: Image compression fundamentals, standards and practice. Kluwer Academic Publishers Norwell, MA, [3] M. Elad, Sparse and redundant representations - From theory to applications in signal and image processing. Springer, [4] S. Mallat and Z. Zhang, Matching pursuits with time-frequency dictionaries, IEEE Trans. Signal Process., vol. 41, no. 12, pp , [5] Y. C. Pati, R. Rezaiifar, and P. S. Krishnaprasad, Orthogonal matching pursuit: recursive function approximation with applications to wavelet decomposition, 1993 Conference Record of The 27th Asilomar Conference on Signals, Systems and Computers, pp , [6] S. S. Chen, D. L. Donoho, and M. A. Saunders, Atomic decomposition by basis pursuit, Technical Report Statistics, Stanford, [7] O. Bryt and M. Elad, Compression of facial images using the K-SVD algorithm, Journal of Visual Communication and Image Representation, vol. 19, no. 4, pp , [8] K. S. Gurumoorthy, A. Rajwade, A. Banerjee, and A. Rangarajan, A method for compact image representation using sparse matrix and tensor projections onto exemplar orthonormal bases, IEEE Trans. Image Process., vol. 19, no. 2, pp , [9] J. Zepeda, C. Guillemot, and E. Kijak, Image compression using sparse representations and the iteration-tuned and aligned dictionary, Submitted. [10] R. Rubinstein, A. M. Bruckstein, and M. Elad, Dictionaries for sparse representation modeling, IEEE Proc., vol. 98, no. 6, pp , [11] R. Rubinstein, M. Zibulevsky, and M. Elad, Double sparsity: learning sparse dictionaries for sparse signal approximation, IEEE Trans. Signal Process., vol. 58, no. 3, pp , [12] M. Aharon, M. Elad, and A. M. Bruckstein, The K-SVD: an algorithm for designing of overcomplete dictionaries for sparse representation, IEEE Trans. Signal Process., vol. 54, no. 11, pp , [13] B. Ophir and M. Elad, Interpolated dictionaries a multi-scale dictionary learning approach, Submitted. [14] I. Horev, Adaptive image compression using Sparse K-SVD, Project Report, Signal and Image Processing Lab (SIPL), EE Technion,

Department of Electronics and Communication KMP College of Engineering, Perumbavoor, Kerala, India 1 2

Department of Electronics and Communication KMP College of Engineering, Perumbavoor, Kerala, India 1 2 Vol.3, Issue 3, 2015, Page.1115-1021 Effect of Anti-Forensics and Dic.TV Method for Reducing Artifact in JPEG Decompression 1 Deepthy Mohan, 2 Sreejith.H 1 PG Scholar, 2 Assistant Professor Department

More information

Efficient Implementation of the K-SVD Algorithm and the Batch-OMP Method

Efficient Implementation of the K-SVD Algorithm and the Batch-OMP Method Efficient Implementation of the K-SVD Algorithm and the Batch-OMP Method Ron Rubinstein, Michael Zibulevsky and Michael Elad Abstract The K-SVD algorithm is a highly effective method of training overcomplete

More information

Efficient Implementation of the K-SVD Algorithm using Batch Orthogonal Matching Pursuit

Efficient Implementation of the K-SVD Algorithm using Batch Orthogonal Matching Pursuit Efficient Implementation of the K-SVD Algorithm using Batch Orthogonal Matching Pursuit Ron Rubinstein, Michael Zibulevsky and Michael Elad Abstract The K-SVD algorithm is a highly effective method of

More information

Image Processing Via Pixel Permutations

Image Processing Via Pixel Permutations Image Processing Via Pixel Permutations Michael Elad The Computer Science Department The Technion Israel Institute of technology Haifa 32000, Israel Joint work with Idan Ram Israel Cohen The Electrical

More information

MRT based Fixed Block size Transform Coding

MRT based Fixed Block size Transform Coding 3 MRT based Fixed Block size Transform Coding Contents 3.1 Transform Coding..64 3.1.1 Transform Selection...65 3.1.2 Sub-image size selection... 66 3.1.3 Bit Allocation.....67 3.2 Transform coding using

More information

Stereo Image Compression

Stereo Image Compression Stereo Image Compression Deepa P. Sundar, Debabrata Sengupta, Divya Elayakumar {deepaps, dsgupta, divyae}@stanford.edu Electrical Engineering, Stanford University, CA. Abstract In this report we describe

More information

Scalable Coding of Image Collections with Embedded Descriptors

Scalable Coding of Image Collections with Embedded Descriptors Scalable Coding of Image Collections with Embedded Descriptors N. Adami, A. Boschetti, R. Leonardi, P. Migliorati Department of Electronic for Automation, University of Brescia Via Branze, 38, Brescia,

More information

ECE 533 Digital Image Processing- Fall Group Project Embedded Image coding using zero-trees of Wavelet Transform

ECE 533 Digital Image Processing- Fall Group Project Embedded Image coding using zero-trees of Wavelet Transform ECE 533 Digital Image Processing- Fall 2003 Group Project Embedded Image coding using zero-trees of Wavelet Transform Harish Rajagopal Brett Buehl 12/11/03 Contributions Tasks Harish Rajagopal (%) Brett

More information

A Image Comparative Study using DCT, Fast Fourier, Wavelet Transforms and Huffman Algorithm

A Image Comparative Study using DCT, Fast Fourier, Wavelet Transforms and Huffman Algorithm International Journal of Engineering Research and General Science Volume 3, Issue 4, July-August, 15 ISSN 91-2730 A Image Comparative Study using DCT, Fast Fourier, Wavelet Transforms and Huffman Algorithm

More information

Multi-polarimetric SAR image compression based on sparse representation

Multi-polarimetric SAR image compression based on sparse representation . RESEARCH PAPER. Special Issue SCIENCE CHINA Information Sciences August 2012 Vol. 55 No. 8: 1888 1897 doi: 10.1007/s11432-012-4612-9 Multi-polarimetric SAR based on sparse representation CHEN Yuan, ZHANG

More information

Generalized Tree-Based Wavelet Transform and Applications to Patch-Based Image Processing

Generalized Tree-Based Wavelet Transform and Applications to Patch-Based Image Processing Generalized Tree-Based Wavelet Transform and * Michael Elad The Computer Science Department The Technion Israel Institute of technology Haifa 32000, Israel *Joint work with A Seminar in the Hebrew University

More information

Key words: B- Spline filters, filter banks, sub band coding, Pre processing, Image Averaging IJSER

Key words: B- Spline filters, filter banks, sub band coding, Pre processing, Image Averaging IJSER International Journal of Scientific & Engineering Research, Volume 7, Issue 9, September-2016 470 Analyzing Low Bit Rate Image Compression Using Filters and Pre Filtering PNV ABHISHEK 1, U VINOD KUMAR

More information

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING

IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING IMAGE DENOISING USING NL-MEANS VIA SMOOTH PATCH ORDERING Idan Ram, Michael Elad and Israel Cohen Department of Electrical Engineering Department of Computer Science Technion - Israel Institute of Technology

More information

ISSN (ONLINE): , VOLUME-3, ISSUE-1,

ISSN (ONLINE): , VOLUME-3, ISSUE-1, PERFORMANCE ANALYSIS OF LOSSLESS COMPRESSION TECHNIQUES TO INVESTIGATE THE OPTIMUM IMAGE COMPRESSION TECHNIQUE Dr. S. Swapna Rani Associate Professor, ECE Department M.V.S.R Engineering College, Nadergul,

More information

Image Restoration and Background Separation Using Sparse Representation Framework

Image Restoration and Background Separation Using Sparse Representation Framework Image Restoration and Background Separation Using Sparse Representation Framework Liu, Shikun Abstract In this paper, we introduce patch-based PCA denoising and k-svd dictionary learning method for the

More information

PATCH-DISAGREEMENT AS A WAY TO IMPROVE K-SVD DENOISING

PATCH-DISAGREEMENT AS A WAY TO IMPROVE K-SVD DENOISING PATCH-DISAGREEMENT AS A WAY TO IMPROVE K-SVD DENOISING Yaniv Romano Department of Electrical Engineering Technion, Haifa 32000, Israel yromano@tx.technion.ac.il Michael Elad Department of Computer Science

More information

arxiv: v1 [cs.cv] 7 Jul 2017

arxiv: v1 [cs.cv] 7 Jul 2017 A MULTI-LAYER IMAGE REPRESENTATION USING REGULARIZED RESIDUAL QUANTIZATION: APPLICATION TO COMPRESSION AND DENOISING Sohrab Ferdowsi, Slava Voloshynovskiy, Dimche Kostadinov Department of Computer Science,

More information

Vidhya.N.S. Murthy Student I.D Project report for Multimedia Processing course (EE5359) under Dr. K.R. Rao

Vidhya.N.S. Murthy Student I.D Project report for Multimedia Processing course (EE5359) under Dr. K.R. Rao STUDY AND IMPLEMENTATION OF THE MATCHING PURSUIT ALGORITHM AND QUALITY COMPARISON WITH DISCRETE COSINE TRANSFORM IN AN MPEG2 ENCODER OPERATING AT LOW BITRATES Vidhya.N.S. Murthy Student I.D. 1000602564

More information

Image Compression for Mobile Devices using Prediction and Direct Coding Approach

Image Compression for Mobile Devices using Prediction and Direct Coding Approach Image Compression for Mobile Devices using Prediction and Direct Coding Approach Joshua Rajah Devadason M.E. scholar, CIT Coimbatore, India Mr. T. Ramraj Assistant Professor, CIT Coimbatore, India Abstract

More information

Overcompressing JPEG images with Evolution Algorithms

Overcompressing JPEG images with Evolution Algorithms Author manuscript, published in "EvoIASP2007, Valencia : Spain (2007)" Overcompressing JPEG images with Evolution Algorithms Jacques Lévy Véhel 1, Franklin Mendivil 2 and Evelyne Lutton 1 1 Inria, Complex

More information

Image Deblurring Using Adaptive Sparse Domain Selection and Adaptive Regularization

Image Deblurring Using Adaptive Sparse Domain Selection and Adaptive Regularization Volume 3, No. 3, May-June 2012 International Journal of Advanced Research in Computer Science RESEARCH PAPER Available Online at www.ijarcs.info ISSN No. 0976-5697 Image Deblurring Using Adaptive Sparse

More information

COMPRESSIVE VIDEO SAMPLING

COMPRESSIVE VIDEO SAMPLING COMPRESSIVE VIDEO SAMPLING Vladimir Stanković and Lina Stanković Dept of Electronic and Electrical Engineering University of Strathclyde, Glasgow, UK phone: +44-141-548-2679 email: {vladimir,lina}.stankovic@eee.strath.ac.uk

More information

The Encoding Complexity of Network Coding

The Encoding Complexity of Network Coding The Encoding Complexity of Network Coding Michael Langberg Alexander Sprintson Jehoshua Bruck California Institute of Technology Email: mikel,spalex,bruck @caltech.edu Abstract In the multicast network

More information

Modified SPIHT Image Coder For Wireless Communication

Modified SPIHT Image Coder For Wireless Communication Modified SPIHT Image Coder For Wireless Communication M. B. I. REAZ, M. AKTER, F. MOHD-YASIN Faculty of Engineering Multimedia University 63100 Cyberjaya, Selangor Malaysia Abstract: - The Set Partitioning

More information

INCOHERENT DICTIONARY LEARNING FOR SPARSE REPRESENTATION BASED IMAGE DENOISING

INCOHERENT DICTIONARY LEARNING FOR SPARSE REPRESENTATION BASED IMAGE DENOISING INCOHERENT DICTIONARY LEARNING FOR SPARSE REPRESENTATION BASED IMAGE DENOISING Jin Wang 1, Jian-Feng Cai 2, Yunhui Shi 1 and Baocai Yin 1 1 Beijing Key Laboratory of Multimedia and Intelligent Software

More information

MRT based Adaptive Transform Coder with Classified Vector Quantization (MATC-CVQ)

MRT based Adaptive Transform Coder with Classified Vector Quantization (MATC-CVQ) 5 MRT based Adaptive Transform Coder with Classified Vector Quantization (MATC-CVQ) Contents 5.1 Introduction.128 5.2 Vector Quantization in MRT Domain Using Isometric Transformations and Scaling.130 5.2.1

More information

Image Inpainting Using Sparsity of the Transform Domain

Image Inpainting Using Sparsity of the Transform Domain Image Inpainting Using Sparsity of the Transform Domain H. Hosseini*, N.B. Marvasti, Student Member, IEEE, F. Marvasti, Senior Member, IEEE Advanced Communication Research Institute (ACRI) Department of

More information

SHIP WAKE DETECTION FOR SAR IMAGES WITH COMPLEX BACKGROUNDS BASED ON MORPHOLOGICAL DICTIONARY LEARNING

SHIP WAKE DETECTION FOR SAR IMAGES WITH COMPLEX BACKGROUNDS BASED ON MORPHOLOGICAL DICTIONARY LEARNING SHIP WAKE DETECTION FOR SAR IMAGES WITH COMPLEX BACKGROUNDS BASED ON MORPHOLOGICAL DICTIONARY LEARNING Guozheng Yang 1, 2, Jing Yu 3, Chuangbai Xiao 3, Weidong Sun 1 1 State Key Laboratory of Intelligent

More information

Wavelet Based Image Compression Using ROI SPIHT Coding

Wavelet Based Image Compression Using ROI SPIHT Coding International Journal of Information & Computation Technology. ISSN 0974-2255 Volume 1, Number 2 (2011), pp. 69-76 International Research Publications House http://www.irphouse.com Wavelet Based Image

More information

Video Compression Method for On-Board Systems of Construction Robots

Video Compression Method for On-Board Systems of Construction Robots Video Compression Method for On-Board Systems of Construction Robots Andrei Petukhov, Michael Rachkov Moscow State Industrial University Department of Automatics, Informatics and Control Systems ul. Avtozavodskaya,

More information

Image Restoration Using DNN

Image Restoration Using DNN Image Restoration Using DNN Hila Levi & Eran Amar Images were taken from: http://people.tuebingen.mpg.de/burger/neural_denoising/ Agenda Domain Expertise vs. End-to-End optimization Image Denoising and

More information

MEMORY EFFICIENT WDR (WAVELET DIFFERENCE REDUCTION) using INVERSE OF ECHELON FORM by EQUATION SOLVING

MEMORY EFFICIENT WDR (WAVELET DIFFERENCE REDUCTION) using INVERSE OF ECHELON FORM by EQUATION SOLVING Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC Vol. 3 Issue. 7 July 2014 pg.512

More information

Fingerprint Image Compression

Fingerprint Image Compression Fingerprint Image Compression Ms.Mansi Kambli 1*,Ms.Shalini Bhatia 2 * Student 1*, Professor 2 * Thadomal Shahani Engineering College * 1,2 Abstract Modified Set Partitioning in Hierarchical Tree with

More information

High flexibility scalable image coding

High flexibility scalable image coding High flexibility scalable image coding Pascal Frossard, Pierre Vandergheynst and Rosa M. Figueras i Ventura Signal Processing Institute Swiss Federal Institute of Technology, Lausanne, Switzerland ABSTRACT

More information

CHAPTER 6. 6 Huffman Coding Based Image Compression Using Complex Wavelet Transform. 6.3 Wavelet Transform based compression technique 106

CHAPTER 6. 6 Huffman Coding Based Image Compression Using Complex Wavelet Transform. 6.3 Wavelet Transform based compression technique 106 CHAPTER 6 6 Huffman Coding Based Image Compression Using Complex Wavelet Transform Page No 6.1 Introduction 103 6.2 Compression Techniques 104 103 6.2.1 Lossless compression 105 6.2.2 Lossy compression

More information

Compression of Stereo Images using a Huffman-Zip Scheme

Compression of Stereo Images using a Huffman-Zip Scheme Compression of Stereo Images using a Huffman-Zip Scheme John Hamann, Vickey Yeh Department of Electrical Engineering, Stanford University Stanford, CA 94304 jhamann@stanford.edu, vickey@stanford.edu Abstract

More information

Image Denoising Via Learned Dictionaries and Sparse representation

Image Denoising Via Learned Dictionaries and Sparse representation Image Denoising Via Learned Dictionaries and Sparse representation Michael Elad Michal Aharon Department of Computer Science The Technion - Israel Institute of Technology, Haifa 32 Israel Abstract We address

More information

PRINCIPAL COMPONENT ANALYSIS IMAGE DENOISING USING LOCAL PIXEL GROUPING

PRINCIPAL COMPONENT ANALYSIS IMAGE DENOISING USING LOCAL PIXEL GROUPING PRINCIPAL COMPONENT ANALYSIS IMAGE DENOISING USING LOCAL PIXEL GROUPING Divesh Kumar 1 and Dheeraj Kalra 2 1 Department of Electronics & Communication Engineering, IET, GLA University, Mathura 2 Department

More information

Review and Implementation of DWT based Scalable Video Coding with Scalable Motion Coding.

Review and Implementation of DWT based Scalable Video Coding with Scalable Motion Coding. Project Title: Review and Implementation of DWT based Scalable Video Coding with Scalable Motion Coding. Midterm Report CS 584 Multimedia Communications Submitted by: Syed Jawwad Bukhari 2004-03-0028 About

More information

FAST AND EFFICIENT SPATIAL SCALABLE IMAGE COMPRESSION USING WAVELET LOWER TREES

FAST AND EFFICIENT SPATIAL SCALABLE IMAGE COMPRESSION USING WAVELET LOWER TREES FAST AND EFFICIENT SPATIAL SCALABLE IMAGE COMPRESSION USING WAVELET LOWER TREES J. Oliver, Student Member, IEEE, M. P. Malumbres, Member, IEEE Department of Computer Engineering (DISCA) Technical University

More information

Express Letters. A Simple and Efficient Search Algorithm for Block-Matching Motion Estimation. Jianhua Lu and Ming L. Liou

Express Letters. A Simple and Efficient Search Algorithm for Block-Matching Motion Estimation. Jianhua Lu and Ming L. Liou IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 7, NO. 2, APRIL 1997 429 Express Letters A Simple and Efficient Search Algorithm for Block-Matching Motion Estimation Jianhua Lu and

More information

Efficient Dictionary Based Video Coding with Reduced Side Information

Efficient Dictionary Based Video Coding with Reduced Side Information MITSUBISHI ELECTRIC RESEARCH LABORATORIES http://www.merl.com Efficient Dictionary Based Video Coding with Reduced Side Information Kang, J-W.; Kuo, C.C. J.; Cohen, R.; Vetro, A. TR2011-026 May 2011 Abstract

More information

Structurally Random Matrices

Structurally Random Matrices Fast Compressive Sampling Using Structurally Random Matrices Presented by: Thong Do (thongdo@jhu.edu) The Johns Hopkins University A joint work with Prof. Trac Tran, The Johns Hopkins University it Dr.

More information

Compression of RADARSAT Data with Block Adaptive Wavelets Abstract: 1. Introduction

Compression of RADARSAT Data with Block Adaptive Wavelets Abstract: 1. Introduction Compression of RADARSAT Data with Block Adaptive Wavelets Ian Cumming and Jing Wang Department of Electrical and Computer Engineering The University of British Columbia 2356 Main Mall, Vancouver, BC, Canada

More information

Compressive Sensing for Multimedia. Communications in Wireless Sensor Networks

Compressive Sensing for Multimedia. Communications in Wireless Sensor Networks Compressive Sensing for Multimedia 1 Communications in Wireless Sensor Networks Wael Barakat & Rabih Saliba MDDSP Project Final Report Prof. Brian L. Evans May 9, 2008 Abstract Compressive Sensing is an

More information

QUANTIZER DESIGN FOR EXPLOITING COMMON INFORMATION IN LAYERED CODING. Mehdi Salehifar, Tejaswi Nanjundaswamy, and Kenneth Rose

QUANTIZER DESIGN FOR EXPLOITING COMMON INFORMATION IN LAYERED CODING. Mehdi Salehifar, Tejaswi Nanjundaswamy, and Kenneth Rose QUANTIZER DESIGN FOR EXPLOITING COMMON INFORMATION IN LAYERED CODING Mehdi Salehifar, Tejaswi Nanjundaswamy, and Kenneth Rose Department of Electrical and Computer Engineering University of California,

More information

Image Interpolation using Collaborative Filtering

Image Interpolation using Collaborative Filtering Image Interpolation using Collaborative Filtering 1,2 Qiang Guo, 1,2,3 Caiming Zhang *1 School of Computer Science and Technology, Shandong Economic University, Jinan, 250014, China, qguo2010@gmail.com

More information

SIGNAL COMPRESSION. 9. Lossy image compression: SPIHT and S+P

SIGNAL COMPRESSION. 9. Lossy image compression: SPIHT and S+P SIGNAL COMPRESSION 9. Lossy image compression: SPIHT and S+P 9.1 SPIHT embedded coder 9.2 The reversible multiresolution transform S+P 9.3 Error resilience in embedded coding 178 9.1 Embedded Tree-Based

More information

Optimized Progressive Coding of Stereo Images Using Discrete Wavelet Transform

Optimized Progressive Coding of Stereo Images Using Discrete Wavelet Transform Optimized Progressive Coding of Stereo Images Using Discrete Wavelet Transform Torsten Palfner, Alexander Mali and Erika Müller Institute of Telecommunications and Information Technology, University of

More information

Sparse Models in Image Understanding And Computer Vision

Sparse Models in Image Understanding And Computer Vision Sparse Models in Image Understanding And Computer Vision Jayaraman J. Thiagarajan Arizona State University Collaborators Prof. Andreas Spanias Karthikeyan Natesan Ramamurthy Sparsity Sparsity of a vector

More information

A parallel patch based algorithm for CT image denoising on the Cell Broadband Engine

A parallel patch based algorithm for CT image denoising on the Cell Broadband Engine A parallel patch based algorithm for CT image denoising on the Cell Broadband Engine Dominik Bartuschat, Markus Stürmer, Harald Köstler and Ulrich Rüde Friedrich-Alexander Universität Erlangen-Nürnberg,Germany

More information

ANALYSIS OF SPIHT ALGORITHM FOR SATELLITE IMAGE COMPRESSION

ANALYSIS OF SPIHT ALGORITHM FOR SATELLITE IMAGE COMPRESSION ANALYSIS OF SPIHT ALGORITHM FOR SATELLITE IMAGE COMPRESSION K Nagamani (1) and AG Ananth (2) (1) Assistant Professor, R V College of Engineering, Bangalore-560059. knmsm_03@yahoo.com (2) Professor, R V

More information

1740 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 19, NO. 7, JULY /$ IEEE

1740 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 19, NO. 7, JULY /$ IEEE 1740 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 19, NO. 7, JULY 2010 Direction-Adaptive Partitioned Block Transform for Color Image Coding Chuo-Ling Chang, Member, IEEE, Mina Makar, Student Member, IEEE,

More information

International Journal of Emerging Technology and Advanced Engineering Website: (ISSN , Volume 2, Issue 4, April 2012)

International Journal of Emerging Technology and Advanced Engineering Website:   (ISSN , Volume 2, Issue 4, April 2012) A Technical Analysis Towards Digital Video Compression Rutika Joshi 1, Rajesh Rai 2, Rajesh Nema 3 1 Student, Electronics and Communication Department, NIIST College, Bhopal, 2,3 Prof., Electronics and

More information

Compression of Light Field Images using Projective 2-D Warping method and Block matching

Compression of Light Field Images using Projective 2-D Warping method and Block matching Compression of Light Field Images using Projective 2-D Warping method and Block matching A project Report for EE 398A Anand Kamat Tarcar Electrical Engineering Stanford University, CA (anandkt@stanford.edu)

More information

Image Compression Algorithm and JPEG Standard

Image Compression Algorithm and JPEG Standard International Journal of Scientific and Research Publications, Volume 7, Issue 12, December 2017 150 Image Compression Algorithm and JPEG Standard Suman Kunwar sumn2u@gmail.com Summary. The interest in

More information

On Single Image Scale-Up using Sparse-Representation

On Single Image Scale-Up using Sparse-Representation On Single Image Scale-Up using Sparse-Representation Roman Zeyde, Matan Protter and Michael Elad The Computer Science Department Technion Israel Institute of Technology Haifa 32000, Israel {romanz,matanpr,elad}@cs.technion.ac.il

More information

A new predictive image compression scheme using histogram analysis and pattern matching

A new predictive image compression scheme using histogram analysis and pattern matching University of Wollongong Research Online University of Wollongong in Dubai - Papers University of Wollongong in Dubai 00 A new predictive image compression scheme using histogram analysis and pattern matching

More information

Image Denoising Using Sparse Representations

Image Denoising Using Sparse Representations Image Denoising Using Sparse Representations SeyyedMajid Valiollahzadeh 1,,HamedFirouzi 1, Massoud Babaie-Zadeh 1, and Christian Jutten 2 1 Department of Electrical Engineering, Sharif University of Technology,

More information

Comparison of EBCOT Technique Using HAAR Wavelet and Hadamard Transform

Comparison of EBCOT Technique Using HAAR Wavelet and Hadamard Transform Comparison of EBCOT Technique Using HAAR Wavelet and Hadamard Transform S. Aruna Deepthi, Vibha D. Kulkarni, Dr.K. Jaya Sankar Department of Electronics and Communication Engineering, Vasavi College of

More information

Rate-distortion Optimized Streaming of Compressed Light Fields with Multiple Representations

Rate-distortion Optimized Streaming of Compressed Light Fields with Multiple Representations Rate-distortion Optimized Streaming of Compressed Light Fields with Multiple Representations Prashant Ramanathan and Bernd Girod Department of Electrical Engineering Stanford University Stanford CA 945

More information

BSIK-SVD: A DICTIONARY-LEARNING ALGORITHM FOR BLOCK-SPARSE REPRESENTATIONS. Yongqin Zhang, Jiaying Liu, Mading Li, Zongming Guo

BSIK-SVD: A DICTIONARY-LEARNING ALGORITHM FOR BLOCK-SPARSE REPRESENTATIONS. Yongqin Zhang, Jiaying Liu, Mading Li, Zongming Guo 2014 IEEE International Conference on Acoustic, Speech and Signal Processing (ICASSP) BSIK-SVD: A DICTIONARY-LEARNING ALGORITHM FOR BLOCK-SPARSE REPRESENTATIONS Yongqin Zhang, Jiaying Liu, Mading Li, Zongming

More information

AN ANALYTICAL STUDY OF LOSSY COMPRESSION TECHINIQUES ON CONTINUOUS TONE GRAPHICAL IMAGES

AN ANALYTICAL STUDY OF LOSSY COMPRESSION TECHINIQUES ON CONTINUOUS TONE GRAPHICAL IMAGES AN ANALYTICAL STUDY OF LOSSY COMPRESSION TECHINIQUES ON CONTINUOUS TONE GRAPHICAL IMAGES Dr.S.Narayanan Computer Centre, Alagappa University, Karaikudi-South (India) ABSTRACT The programs using complex

More information

CSEP 521 Applied Algorithms Spring Lossy Image Compression

CSEP 521 Applied Algorithms Spring Lossy Image Compression CSEP 521 Applied Algorithms Spring 2005 Lossy Image Compression Lossy Image Compression Methods Scalar quantization (SQ). Vector quantization (VQ). DCT Compression JPEG Wavelet Compression SPIHT UWIC (University

More information

Embedded Rate Scalable Wavelet-Based Image Coding Algorithm with RPSWS

Embedded Rate Scalable Wavelet-Based Image Coding Algorithm with RPSWS Embedded Rate Scalable Wavelet-Based Image Coding Algorithm with RPSWS Farag I. Y. Elnagahy Telecommunications Faculty of Electrical Engineering Czech Technical University in Prague 16627, Praha 6, Czech

More information

Comparative Analysis of 2-Level and 4-Level DWT for Watermarking and Tampering Detection

Comparative Analysis of 2-Level and 4-Level DWT for Watermarking and Tampering Detection International Journal of Latest Engineering and Management Research (IJLEMR) ISSN: 2455-4847 Volume 1 Issue 4 ǁ May 2016 ǁ PP.01-07 Comparative Analysis of 2-Level and 4-Level for Watermarking and Tampering

More information

A Novel Approach for Image Compression using Matching Pursuit Signal Approximation and Simulated Annealing

A Novel Approach for Image Compression using Matching Pursuit Signal Approximation and Simulated Annealing A Novel Approach for Image Compression using Matching Pursuit Signal Approximation and Simulated Annealing Ahmed M. Eid Amin ahmedamin@ieee.org Supervised by: Prof. Dr. Samir Shaheen Prof. Dr. Amir Atiya

More information

Color Image Compression Using EZW and SPIHT Algorithm

Color Image Compression Using EZW and SPIHT Algorithm Color Image Compression Using EZW and SPIHT Algorithm Ms. Swati Pawar 1, Mrs. Adita Nimbalkar 2, Mr. Vivek Ugale 3 swati.pawar@sitrc.org 1, adita.nimbalkar@sitrc.org 2, vivek.ugale@sitrc.org 3 Department

More information

Packed Integer Wavelet Transform Constructed by Lifting Scheme

Packed Integer Wavelet Transform Constructed by Lifting Scheme 1496 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 10, NO. 8, DECEMBER 2000 Packed Integer Wavelet Transform Constructed by Lting Scheme Chengjiang Lin, Bo Zhang, and Yuan F. Zheng

More information

An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising

An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising An Optimized Pixel-Wise Weighting Approach For Patch-Based Image Denoising Dr. B. R.VIKRAM M.E.,Ph.D.,MIEEE.,LMISTE, Principal of Vijay Rural Engineering College, NIZAMABAD ( Dt.) G. Chaitanya M.Tech,

More information

An Approach for Reduction of Rain Streaks from a Single Image

An Approach for Reduction of Rain Streaks from a Single Image An Approach for Reduction of Rain Streaks from a Single Image Vijayakumar Majjagi 1, Netravati U M 2 1 4 th Semester, M. Tech, Digital Electronics, Department of Electronics and Communication G M Institute

More information

A NEW ROBUST IMAGE WATERMARKING SCHEME BASED ON DWT WITH SVD

A NEW ROBUST IMAGE WATERMARKING SCHEME BASED ON DWT WITH SVD A NEW ROBUST IMAGE WATERMARKING SCHEME BASED ON WITH S.Shanmugaprabha PG Scholar, Dept of Computer Science & Engineering VMKV Engineering College, Salem India N.Malmurugan Director Sri Ranganathar Institute

More information

CMPT 365 Multimedia Systems. Media Compression - Image

CMPT 365 Multimedia Systems. Media Compression - Image CMPT 365 Multimedia Systems Media Compression - Image Spring 2017 Edited from slides by Dr. Jiangchuan Liu CMPT365 Multimedia Systems 1 Facts about JPEG JPEG - Joint Photographic Experts Group International

More information

Image Compression. CS 6640 School of Computing University of Utah

Image Compression. CS 6640 School of Computing University of Utah Image Compression CS 6640 School of Computing University of Utah Compression What Reduce the amount of information (bits) needed to represent image Why Transmission Storage Preprocessing Redundant & Irrelevant

More information

CHAPTER 9 INPAINTING USING SPARSE REPRESENTATION AND INVERSE DCT

CHAPTER 9 INPAINTING USING SPARSE REPRESENTATION AND INVERSE DCT CHAPTER 9 INPAINTING USING SPARSE REPRESENTATION AND INVERSE DCT 9.1 Introduction In the previous chapters the inpainting was considered as an iterative algorithm. PDE based method uses iterations to converge

More information

University of Mustansiriyah, Baghdad, Iraq

University of Mustansiriyah, Baghdad, Iraq Volume 5, Issue 9, September 2015 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Audio Compression

More information

Robust Image Watermarking based on DCT-DWT- SVD Method

Robust Image Watermarking based on DCT-DWT- SVD Method Robust Image Watermarking based on DCT-DWT- SVD Sneha Jose Rajesh Cherian Roy, PhD. Sreenesh Shashidharan ABSTRACT Hybrid Image watermarking scheme proposed based on Discrete Cosine Transform (DCT)-Discrete

More information

CHAPTER 4 REVERSIBLE IMAGE WATERMARKING USING BIT PLANE CODING AND LIFTING WAVELET TRANSFORM

CHAPTER 4 REVERSIBLE IMAGE WATERMARKING USING BIT PLANE CODING AND LIFTING WAVELET TRANSFORM 74 CHAPTER 4 REVERSIBLE IMAGE WATERMARKING USING BIT PLANE CODING AND LIFTING WAVELET TRANSFORM Many data embedding methods use procedures that in which the original image is distorted by quite a small

More information

Rate-distortion Optimized Streaming of Compressed Light Fields with Multiple Representations

Rate-distortion Optimized Streaming of Compressed Light Fields with Multiple Representations Rate-distortion Optimized Streaming of Compressed Light Fields with Multiple Representations Prashant Ramanathan and Bernd Girod Department of Electrical Engineering Stanford University Stanford CA 945

More information

A Low-power, Low-memory System for Wavelet-based Image Compression

A Low-power, Low-memory System for Wavelet-based Image Compression A Low-power, Low-memory System for Wavelet-based Image Compression James S. Walker Department of Mathematics University of Wisconsin Eau Claire Truong Q. Nguyen Department of Electrical and Computer Engineering

More information

Performance Comparison between DWT-based and DCT-based Encoders

Performance Comparison between DWT-based and DCT-based Encoders , pp.83-87 http://dx.doi.org/10.14257/astl.2014.75.19 Performance Comparison between DWT-based and DCT-based Encoders Xin Lu 1 and Xuesong Jin 2 * 1 School of Electronics and Information Engineering, Harbin

More information

Short Communications

Short Communications Pertanika J. Sci. & Technol. 9 (): 9 35 (0) ISSN: 08-7680 Universiti Putra Malaysia Press Short Communications Singular Value Decomposition Based Sub-band Decomposition and Multiresolution (SVD-SBD-MRR)

More information

The Viterbi Algorithm for Subset Selection

The Viterbi Algorithm for Subset Selection 524 IEEE SIGNAL PROCESSING LETTERS, VOL. 22, NO. 5, MAY 2015 The Viterbi Algorithm for Subset Selection Shay Maymon and Yonina C. Eldar, Fellow, IEEE Abstract We study the problem of sparse recovery in

More information

DIGITAL TELEVISION 1. DIGITAL VIDEO FUNDAMENTALS

DIGITAL TELEVISION 1. DIGITAL VIDEO FUNDAMENTALS DIGITAL TELEVISION 1. DIGITAL VIDEO FUNDAMENTALS Television services in Europe currently broadcast video at a frame rate of 25 Hz. Each frame consists of two interlaced fields, giving a field rate of 50

More information

Keywords DCT, SPIHT, PSNR, Bar Graph, Compression Quality

Keywords DCT, SPIHT, PSNR, Bar Graph, Compression Quality Volume 3, Issue 7, July 2013 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Image Compression

More information

Data Hiding in Video

Data Hiding in Video Data Hiding in Video J. J. Chae and B. S. Manjunath Department of Electrical and Computer Engineering University of California, Santa Barbara, CA 9316-956 Email: chaejj, manj@iplab.ece.ucsb.edu Abstract

More information

Novel Lossy Compression Algorithms with Stacked Autoencoders

Novel Lossy Compression Algorithms with Stacked Autoencoders Novel Lossy Compression Algorithms with Stacked Autoencoders Anand Atreya and Daniel O Shea {aatreya, djoshea}@stanford.edu 11 December 2009 1. Introduction 1.1. Lossy compression Lossy compression is

More information

EFFICIENT DICTIONARY LEARNING IMPLEMENTATION ON THE GPU USING OPENCL

EFFICIENT DICTIONARY LEARNING IMPLEMENTATION ON THE GPU USING OPENCL U.P.B. Sci. Bull., Series C, Vol. 78, Iss. 3, 2016 ISSN 2286-3540 EFFICIENT DICTIONARY LEARNING IMPLEMENTATION ON THE GPU USING OPENCL Paul Irofti 1 Abstract The dictionary learning field offers a wide

More information

Outline Introduction MPEG-2 MPEG-4. Video Compression. Introduction to MPEG. Prof. Pratikgiri Goswami

Outline Introduction MPEG-2 MPEG-4. Video Compression. Introduction to MPEG. Prof. Pratikgiri Goswami to MPEG Prof. Pratikgiri Goswami Electronics & Communication Department, Shree Swami Atmanand Saraswati Institute of Technology, Surat. Outline of Topics 1 2 Coding 3 Video Object Representation Outline

More information

Reconstruction PSNR [db]

Reconstruction PSNR [db] Proc. Vision, Modeling, and Visualization VMV-2000 Saarbrücken, Germany, pp. 199-203, November 2000 Progressive Compression and Rendering of Light Fields Marcus Magnor, Andreas Endmann Telecommunications

More information

Tutorial on Image Compression

Tutorial on Image Compression Tutorial on Image Compression Richard Baraniuk Rice University dsp.rice.edu Agenda Image compression problem Transform coding (lossy) Approximation linear, nonlinear DCT-based compression JPEG Wavelet-based

More information

TERM PAPER ON The Compressive Sensing Based on Biorthogonal Wavelet Basis

TERM PAPER ON The Compressive Sensing Based on Biorthogonal Wavelet Basis TERM PAPER ON The Compressive Sensing Based on Biorthogonal Wavelet Basis Submitted By: Amrita Mishra 11104163 Manoj C 11104059 Under the Guidance of Dr. Sumana Gupta Professor Department of Electrical

More information

Expected Patch Log Likelihood with a Sparse Prior

Expected Patch Log Likelihood with a Sparse Prior Expected Patch Log Likelihood with a Sparse Prior Jeremias Sulam and Michael Elad Computer Science Department, Technion, Israel {jsulam,elad}@cs.technion.ac.il Abstract. Image priors are of great importance

More information

Sparse Modeling of Graph-Structured Data

Sparse Modeling of Graph-Structured Data Sparse Modeling of Graph-Structured Data and Images Michael Elad The Computer Science Department The Technion The 3rd Annual International TCE Conference Machine Learning & Big Data May 28-29, 2013 1 SPARSITY:

More information

Image Interpolation Using Multiscale Geometric Representations

Image Interpolation Using Multiscale Geometric Representations Image Interpolation Using Multiscale Geometric Representations Nickolaus Mueller, Yue Lu and Minh N. Do Department of Electrical and Computer Engineering University of Illinois at Urbana-Champaign ABSTRACT

More information

2386 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 6, JUNE 2006

2386 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 6, JUNE 2006 2386 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 6, JUNE 2006 The Encoding Complexity of Network Coding Michael Langberg, Member, IEEE, Alexander Sprintson, Member, IEEE, and Jehoshua Bruck,

More information

Compressed Sensing Algorithm for Real-Time Doppler Ultrasound Image Reconstruction

Compressed Sensing Algorithm for Real-Time Doppler Ultrasound Image Reconstruction Mathematical Modelling and Applications 2017; 2(6): 75-80 http://www.sciencepublishinggroup.com/j/mma doi: 10.11648/j.mma.20170206.14 ISSN: 2575-1786 (Print); ISSN: 2575-1794 (Online) Compressed Sensing

More information

IMAGE COMPRESSION USING HYBRID QUANTIZATION METHOD IN JPEG

IMAGE COMPRESSION USING HYBRID QUANTIZATION METHOD IN JPEG IMAGE COMPRESSION USING HYBRID QUANTIZATION METHOD IN JPEG MANGESH JADHAV a, SNEHA GHANEKAR b, JIGAR JAIN c a 13/A Krishi Housing Society, Gokhale Nagar, Pune 411016,Maharashtra, India. (mail2mangeshjadhav@gmail.com)

More information

Efficient Image Compression of Medical Images Using the Wavelet Transform and Fuzzy c-means Clustering on Regions of Interest.

Efficient Image Compression of Medical Images Using the Wavelet Transform and Fuzzy c-means Clustering on Regions of Interest. Efficient Image Compression of Medical Images Using the Wavelet Transform and Fuzzy c-means Clustering on Regions of Interest. D.A. Karras, S.A. Karkanis and D. E. Maroulis University of Piraeus, Dept.

More information

Introduction to Video Compression

Introduction to Video Compression Insight, Analysis, and Advice on Signal Processing Technology Introduction to Video Compression Jeff Bier Berkeley Design Technology, Inc. info@bdti.com http://www.bdti.com Outline Motivation and scope

More information